Alexander Larsson
b441dfa729
We make a tmpfs on /sys/fs/cgroups, and here we mount read-only versions of all the host cgroups. Additionally we make symlinks for all merged subsystems. For any "named" cgroup, such as "name=systemd" we also mount the subset of the cgroup where the container lives as read-write. This means that the container can create sub-cgroups inside the container and move tasks into those, but it can never escape from its current position in the cgroup hierarchy. In particular, this allows systemd to mostly work in a non-privileged container. The only problem currently is that PrivateTmp=true fails because systemd is not allowed to mount a new /tmp. Docker-DCO-1.1-Signed-off-by: Alexander Larsson <alexl@redhat.com> (github: alexlarsson)
libcontainer - reference implementation for containers
Background
libcontainer specifies configuration options for what a container is. It provides a native Go implementation for using Linux namespaces with no external dependencies. libcontainer provides many convenience functions for working with namespaces, networking, and management.
Container
A container is a self contained directory that is able to run one or more processes without
affecting the host system. The directory is usually a full system tree. Inside the directory
a container.json file is placed with the runtime configuration for how the processes
should be contained and run. Environment, networking, and different capabilities for the
process are specified in this file. The configuration is used for each process executed inside the container.
See the container.json file for what the configuration should look like.
Using this configuration and the current directory holding the rootfs for a process, one can use libcontainer to exec the container. Running the life of the namespace, a pid file
is written to the current directory with the pid of the namespaced process to the external world. A client can use this pid to wait, kill, or perform other operation with the container. If a user tries to run a new process inside an existing container with a live namespace, the namespace will be joined by the new process.
You may also specify an alternate root place where the container.json file is read and where the pid file will be saved.
nsinit
nsinit is a cli application used as the reference implementation of libcontainer. It is able to
spawn or join new containers giving the current directory. To use nsinit cd into a Linux
rootfs and copy a container.json file into the directory with your specified configuration.
To execute /bin/bash in the current directory as a container just run:
nsinit exec /bin/bash
If you wish to spawn another process inside the container while your current bash session is running just run the exact same command again to get another bash shell or change the command. If the original process dies, PID 1, all other processes spawned inside the container will also be killed and the namespace will be removed.
You can identify if a process is running in a container by looking to see if pid is in the root of the directory.
Future
See the roadmap.
Copyright and license
Code and documentation copyright 2014 Docker, inc. Code released under the Apache 2.0 license. Docs released under Creative commons.
Hacking on libcontainer
First of all, please familiarise yourself with the libcontainer Principles.
If you're a contributor or aspiring contributor, you should read the Contributors' Guide.
If you're a maintainer or aspiring maintainer, you should read the Maintainers' Guide and "How can I become a maintainer?" in the Contributors' Guide.